Why is my new Elasticsearch index slower under load despite optimized mapping?

Question

We've created a new Elasticsearch index (test-events-v4) with what we believed to be a more optimized mapping for storage. However, during load testing, we've observed significantly higher 99th percentile and max latency compared to the old index (test-events-old) — even when querying the exact same data set and using the same query logic.

Here’s an example query used on both indices:

GET /<index>/_search
{
  "from": 0,
  "size": 10,
  "query": {
    "bool": {
      "should": [
        {
          "bool": {
            "filter": [
              { "term": { "status": <value> }},
              { "term": { "addressId": <value> }}
            ]
          }
        }
      ]
    }
  },
  "sort": [
    {
      "timestamp": {
        "order": "desc"
      }
    }
  ]
}

Key differences between new and old index:

Many of the fields like (status, addressId) was of type text + keyword field, I have changed it to "keyword" field.

Also the fields which are not being searched or aggregated I have added this settings:

"index": false,
"doc_values": false

there were some object type fields I have stopped mapping the nested fields there, like before it was:

metadata: {
   "type": {
     // mapping
   },
   "created": { }
   "mode": {}
}

I changed them now to this:

metadata: {
   "type": "object",
   "dynamic": "false"
}

I ran a benchmark after that, I got this result.

Index            99th Percentile Latency    Max Latency
test-events-old       ~640ms                  ~880ms
test-events-new       ~1050ms                 ~6600ms

Although I see my new mapping has improved index creation time and store size, but query time is significantly higher. Although segments count is not much, segment sizes are uniform I have checked.

index           docs.count docs.deleted store.size segments.count indexing.index_total indexing.index_time
test-events-new      15882          0     31.3mb             10                60000                6.1s
test-events-old      15881          0     42.9mb             12                90000               17.6s

Does "enabled": false or "dynamic": "false" on nested fields contribute to latency even if those fields aren’t used in the query? what should I do?